Had I just picked a language and just ran with it, I would be farther down the road than where I am now. So just grab something, sit down, and play with it.
I started tinkering with Prolog as a summer project, and as a side effect, Erlang makes sooo much more sense now. (Erlang was originally implemented in Prolog, and inherited a lot of its syntax & style.) I picked it deliberately because it's quite far from everything I know, and Peter Norvig said a lot of interesting things about it in PAIP. Chatter on the web gives the impression that it's just some weird dead AI language, but I've found it really inspiring. I'm glad I didn't let that discourage me. Even if the only benefit I get from it is the insight into Erlang's design, being able to read Sterling & Shapiro's _The Art of Prolog_* , and finally getting how unification/pattern matching and type inference work, it was time well spent.
* Which is awe-inspiring. I would recommend it regardless of whether you care about Prolog, much like Norvig's PAIP is worth reading even if you don't care about Common Lisp.
[1]: http://www.cs.brown.edu/~sk/Publications/Talks/SwineBeforePe...
While people can argue about the relative merits of Ruby and Python for days, the direct impact of language characteristics on a beginner are likely to be overwhelmed by other issues. For example, if a friend knows one of them quite well, go with that one - he/she can probably answer some questions, and in the short term that's far more important.
The same principle holds for Linux/BSD distributions, version control systems, and so on: If there's not a huge consensus and the differences don't mean much to you, just pick one. You'll be in a better position to make an informed decision later, and you're not stuck with it forever. Time spent trying to decide could instead be put toward learning the basics.
Another example that comes to mind is systems for parsing. There's ANTLR, lex + yacc, packrat parsers/PEGs, Prolog's DCGs, etc. While they have different strengths, any deep comparison will also nonchalantly throw around terms like LALR(1).